Research And Implementation Of Cold Start Problem In Commodity Recommendation System

With the wide application of the network,the e-commerce platform is growing rapidly.The increase in users and goods has brought about the problem of "information overload." In order to solve it,the Recommendation System(hereinafter referred to as RS)came into being.Although the collaborative filtering algorithm is the most used algorithm in RS,it still has problems such as data sparsity and cold start.The recommended effect is severely affected.This paper first describes RS in detail,introduces the collaborative filtering recommendation algorithm and its limitations.Next,an algorithm improvement is implemented for item-based collaborative filtering and user-based collaborative filtering algorithm.Finally,driven by actual demand,the improved algorithm is applied to the commodity recommendation system based on WeChat applet.The main work of the thesis has the following aspects:Firstly,analyzing the problem that Item-CF appears in the cold start of new items,and proposes a method of combining user history preference information and similar items of new items to predict user preferences and recommend new items to users.Since the method combines the user's history and the feature information of the item,the cold start problem of the new item can be effectively solved.Secondly,improving the similarity measure by adding weights to each feature of the item.The weight factor size is determined by calculating the uncertainty of the user for each feature of the item.The uncertainty of the item to the user is calculated by applying the information entropy in the information theory,and the uncertainty of the item feature is calculated by quantifying the feature of the item in the information entropy.In a personalized recommendation system,the less uncertainty about a feature of an item,the higher the importance of the item's characteristics to the user.This method provides a more efficient solution for similarity calculations and can find a more reasonable set of similar items for the item.Thirdly,for the time bottleneck,scalability and accuracy of User-CF,the method of user clustering and feature acquisition is used to recommend user cold start.In the clustering process,due to the lack of scoring information for new users,consider the use of user feature attribute factors.When a new user enters the system,the user is modeled and the nearest neighbor set is found for the new user through the feature attribute similarity between the users and then clustered using the K-means clustering method.When calculating the average score of the commodity,the concept of adding the scoring scale is proposed to make the score more objective and reasonable..Finally.The experimental data set used in this paper is the actual store dataset Istores of the micro store.According to the experimental results: the two improved algorithms in this paper can solve the cold start problem well and improve the accuracy.